The randomized, placebo-controlled trial is often described as the gold standard for research involving humans, primarily due to the minimization of bias through randomization, prospective follow-up, and, where feasible, participant and investigator blinding. Effective allocation concealment, combined with blinding of participants and investigators, aims to minimize bias in patient selection, adherence, and ascertainment of outcomes. When blinding is attempted, it can be compromised by noticeable characteristics of the intervention (e.g. pill taste), by flawed protocol design or execution, by health effects (good or bad), or by participants' attempts to identify their intervention (1-3). Some investigators argu for better reporting of blinding effectiveness and others have devised statistical methods to do so (1, 4-11). Their arguments are countered by the idea that a broader effort is needed to identify any sources of belief-related bias (13), recognizing that even a successfully blinded subject may hold strong beliefs about which intervention they received and its likely efficacy. Unblinding may be just one of several belief-related sources of bias in RCTs, all of which are poorly understood (14, 15). We already know that without allocation concealment and blinding, we have the potential to exaggerate (16-23), or even reverse a trial's conclusions (24); this bias may affect subjective outcomes more than objective ones (20). Building on previous work that revealed specific expectations among participants at the start of a RCT (12), we propose to investigate the biases associated with participants' beliefs during a large, multi-center, randomized, placebo-controlled trial of calcium and vitamin D in colorectal adenoma chemoprevention. We asked 2813 subjects how effective they thought the study treatments were for specified health effects, and which treatment they would prefer if given the choice. At randomization, and at the middle and end of the trial, we asked them to guess which treatment they had been given, and the reasons for their guess. In parallel, we collected longitudinal symptom and adherence data. We now propose to explore how those beliefs affect subsequent reporting of subjective and objective health outcomes, adherence, and attrition, after adjustments for such factors as randomized treatment, prior symptoms, reasons given by participants for their beliefs, demographic and medical factors. This will be a detailed, longitudinal analysis of the impact of individual subjects' expectations, preferences and beliefs about their assigned treatment on health outcomes and adherence during a large RCT. Our goal is to shift the current paradigm away from its focus on unblinding, by describing directly whether subjects' beliefs about the study treatment can generate bias. If so, we will explore strategies to correct or prevent these biases in future trials. Whether negative or positive, our results will significantly advance RCT methodology, and trigger similar research in other trials. There is currently little published research along the lines that we propose, although the potential importance of biases due to expectations and hunches about treatment efficacy and assignment is increasingly being recognized.